Logit Perturbation

Authors

  • Mengyang Li Jiuantianxia Inc., National Center for Applied Mathematics of Tianjin University
  • Fengguang Su National Center for Applied Mathematics of Tianjin University
  • Ou Wu National Center for Applied Mathematics of Tianjin University
  • Ji Zhang The University of Southern Queensland

DOI:

https://doi.org/10.1609/aaai.v36i2.20024

Keywords:

Computer Vision (CV)

Abstract

Features, logits, and labels are the three primary data when a sample passes through a deep neural network. Feature perturbation and label perturbation receive increasing attention in recent years. They have been proven to be useful in various deep learning approaches. For example, (adversarial) feature perturbation can improve the robustness or even generalization capability of learned models. However, limited studies have explicitly explored for the perturbation of logit vectors. This work discusses several existing methods related to logit perturbation. Based on a unified viewpoint between positive/negative data augmentation and loss variations incurred by logit perturbation, a new method is proposed to explicitly learn to perturb logits. A comparative analysis is conducted for the perturbations used in our and existing methods. Extensive experiments on benchmark image classification data sets and their long-tail versions indicated the competitive performance of our learning method. In addition, existing methods can be further improved by utilizing our method.

Downloads

Published

2022-06-28

How to Cite

Li, M., Su, F., Wu, O., & Zhang, J. (2022). Logit Perturbation. Proceedings of the AAAI Conference on Artificial Intelligence, 36(2), 1359-1366. https://doi.org/10.1609/aaai.v36i2.20024

Issue

Section

AAAI Technical Track on Computer Vision II